v0.2.4 — detects an existing agent

What's new:

• Existing-agent detection: connect to a server that already runs Session AI Agent and the app skips setup and opens Manage your agent — status line (running · engine · model), tap-to-copy bot Session ID, and the usual buttons: Change model, Switch engine, View Session ID, Uninstall, Exit
• No more accidental overwrite when you just want to manage an existing agent
• Managing needs only your SSH password; changing model/engine asks for the OpenCode API key first

Test steps: docs/TESTING.md

v0.2.3 — smart model list, safe model changes, auto-reconnect

What's new:

• Model list is now account-accurate: models blocked for your account (plan limits, privacy/data-training models like muse-spark) are hidden with the reason shown; transient failures don't remove models
• Changing the model is safe: if the new model doesn't respond, the app automatically reverts to your previous working model
• Auto-reconnect: if the SSH connection drops while you type (e.g. you switched to the Session app), the app reconnects by itself before continuing
• Copy feedback: tapping the bot Session ID now glows green and fades back
• Engine switching only activates the new engine once it responds

Test steps: docs/TESTING.md

v0.2.2 — reliable bridge download, DeepSeek V4.1 Flash default

What's new:

• Bridge download fixed: downloads to a file with retries and timeouts, verifies the archive before extracting, and retries up to 3 times — the truncated-download failure (gzip: unexpected end of file) can no longer kill an install
• DeepSeek V4.1 Flash is now the default model (upgrade from V4 Flash) in the app and the installer
• Connect screen: the password field is now labelled Root password
• README: video walkthrough row (YouTube icon) and Websplaining channel row

Test steps: docs/TESTING.md

v0.2.1 — OpenClaw on Node 24, engine fallback, compact UI

What's new:

• OpenClaw installs fixed: OpenClaw 2026.9.5+ requires Node >=24.16 — the installer now provisions Node 24 (Hermes installs stay Node-free)
• One-tap engine fallback: if OpenClaw fails you can install Hermes instead (and vice versa)
• Progress bar fixed: monotonic 0-100% matched to the real step order, with an activity spinner
• Compact Done screen: fits on one phone screen — tap-to-copy Session ID, 2x2 manage buttons, no more scrolling for Exit
• Better failure diagnostics (node/npm/registry/memory + npm log tail shown in-app)
• README redesign with links to Kamatera, OpenCode Go, YouTube and the main site

Test steps: docs/TESTING.md

v0.2.0 — faster models, progress bar, manage actions

What's new:

• Model list is much faster (max_tokens=1 probes, 12x parallel, 6s timeout) and cached for 24h with a Refresh button
• Install shows a live 0-100% progress bar with step labels
• Connect auto-advances to the next step
• New management buttons: Change model, Switch engine, View Session ID, Uninstall (with confirmation), Exit
• New launcher icon: mirrored plain Session logo

Test steps: docs/TESTING.md

v0.1.1 — SSH crypto fix

Fixes the Android-only SSH failure (no such algorithm: x25519 for provider bc) by bypassing Android's stripped BC provider and using ECDH instead of curve25519. Also: "Connect" button, README download button.

Manual test steps: docs/TESTING.md

v0.1.0 — first test build

Debug APK for testing the wizard end-to-end on a scratch VPS.

See docs/TESTING.md for the manual test steps.

Known limits: password SSH only; install wizard only (no manage actions yet).